Research Of Speech Perceptual Hashing And Its Application In Search Over Encrypted Speech

In recent years, with the rapid development of Internet technology, the multimedia technology and multimedia data played an increasingly important role in the exchange of information, and other aspects of information storage. The speech as the most convenient multimedia information, the role of it is particularly important in communication. With the advances of computer storage technology, especially the development of cloud storage technology, storage of speech data is also rapidly increase. The network platform has a huge amount of users, in which the audio information is used very frequent, thus, how to ensure the privacy of data security under the premise of a more efficient handling of large scale speech information has become an urgent problem.In this paper, the speech perceptual hashing scheme suitable for large-scale speech processing and its application to the retrieval algorithm over encrypted speech has been researched. The main work is as follows:(1) Existing speech perceptual hashing algorithms extract the features without distinguishing weights between the hash sequences, resulting in lower efficiency in the large-scale speech data applications. This paper presents a speech perceptual hashing algorithm based on the change characteristics of the time and frequency domain. In this paper, the features are extracted both in time domain and frequency domain. The change characteristic of short-time energy is selected as the time domain features, and the change characteristic of Bark domain energy is selected as the frequency domain features. Thus two sequences of perceptual hashing digests are generated. When the perceptual hash values are matched, the time domain hash digest is firstly done to match. If the determination result is not matched, the final result does not match. Otherwise, the frequency domain perceptual hash digest is done to match and the final result is obtained. The experimental result shows that the proposed scheme has good discrimination, strong robustness, and high matching efficiency.(2)The application of speech is very important such as speech orders, forensic evidence, military secrets and etc, if these important information upload the cloud, which is not protected, can easily lead to information disclosure. The front-end encryption of cloud data is an effective way to protect the data, as well as the amount of encrypted data is increasing, bring a lot of difficulties for retrieve. To solve this problem, an encrypted domain speech retrieval program is proposed. First, an appropriate encryption algorithm is used to encrypt the speech, and then a hash sequence is generated and embedded into the encrypted speech as a digital watermark. In the retrieval process, the watermark and perceptual hash index can be matched without decrypting and downloading the encrypted speech. In addition, retrieve from the large scale encrypted domain speech is fast and accurate. The experimental results show that the scheme has a very good recall and precision.